banana vorticism

is a playful reinterpretation of “Exquisite Corpse” that leverages on the discrepancies of Machine Learning to create unexpected inspiration between machine and artists

✦

Practise : Machine Learning

Timeline: 1 week

Project team: Julian | Kosuke | Shin

Roles: Concept, narrative, protoyping

the context

We were at the start of the pandemic lockdown during our machine learning course. Thus, we took the chance to explore the possibilities of using machine learning to make our lockdown situation better.

How 'machines' could help us make better our present situation of social distancing, remote communication and longer time spent in our home? 

Uhh, can this be more interesting so it gives us ideas?

During brainstorming, the team felt uninspired and unmotivated when we heard about the Covid lockdown. We also learned from our artist friends that they shared same feeling lacking inspirations since the lockdown happened.

We started observing our mundane surrounding and wondered if there's something can be done about it.

eureka moment

eureka moment

eureka moment

The fun part came when we started feeding some mundane imagery to RunwayML to test one of the model and it gave amusingly unexpected result.

We were intrigued by the dicrepencies of RunwayML modal and used it as an advantage to prompt unexpected angle of inspirations. The team decided to push it further.

RunwayML experiments

We put together few combinations of RunwayML models that very loosely determine the type of text and images to be generated. The user receives both the result for each ML model and the final result of the long succesion of ML models that were put together in a sequence.

Camera input

what will this be if translated into text?

im2text model

a desk with a keyboard mouse and a monitor

okaay interesting, let's elaborate this in a poetic way

GTP 2 - Little women model

now let's see what does this prose look like

AttnGAN model

Let's reconstruct for a more vivid imagery

BibiGAN model

This looks like something, let's push for more clarity!

BibiGAN model

interessting....

These are outputs of various combinations

Key reflections

The team loved how the result generally had an expressionist art-vibe to it, it's as if witnessing how the machine "expressing" it's artisic vision in the most uncanny way.

Through the whole process we realised the best way to enjoy learning Machine Learning is to let go of the expectation of an ideal output and fully embrace the machine's perception. 

We even adopted the random name that RunwayML program generated for us when we started running the first project model - Banana Vorticism. We love how it fits well with the context of this experiment - the marriage of machine and urban.

The ML aesthetics may be too avant garde for our normal aesthetic standards, but we learned that Machine Learning is definitely a friend rather than a foe, if we can see it beyond a tool to correct/manipulate human's behaviour, and definitely a fun lens to perceive our world.